In an Atlanta lab, minibrains in dishes can control robots and computer-simulated animals. They may provide a simple model to study how the brain changes as it learns. Georgia Tech researcher Steve Potter calls his biological-mechanical hybrids Neurally Controlled Animats. Each Animat's "brain" consists of cultured rat neurons growing on a plate of electrodes. This live culture is linked to an artificial body—either a robotic animal or a computer-simulated one.
The brain cells can receive input from the synthetic critter's environment, process information, and stimulate behavior. For example, one of the Animats is linked to a robot with light-detecting sensors. Information from the sensors is sent to the brain culture, which sends a signal back to the robot instructing it to move toward the light. Other Animats can chase a target around a room, scribble simple drawings, and move around obstacles.
Until recently, training Animats had been tough. The cultured brain cells kept firing in synchronized bursts that disrupted learning. Potter suspected this bursting was a symptom of sensory deprivation, since the neurons weren't receiving constant electrical input as they would in a living body. He was able to quiet the neural fireworks by applying soothing background stimulation. "Now we've seen reliably detectable changes. We think we're on the right track." Animats won't be taking over the world anytime soon, but Potter acknowledges that these "semiliving objects" are smarter than your average petri dish. "They can process information from their environment and respond to it. It's not human consciousness, but it's something."
